Website visuals, from first principles

Maps to: “Getting every system to agree about every sale”

Every system agrees

Settlements tie out to the ledger; matches pass silently and dim, the one mismatch stays lit as a same-day exception with its records attached.

connector revealexception pulse

Maps to: reconciliation root-cause analysis

Where the gap was hiding

A recovered reconciliation gap, broken to root cause in four bars that grow on the GPU with rand values counting up in sync.

scaleX staggercount-up

15 lines JS1.8 KB gzip

Who wins, who gives back

A candidate commission scheme replayed on real transactions — each rep's pay slope named as a gain (teal) or a give-back (amber) before anyone's pay changes.

Maps to: “Redesigning a sales commission scheme”

slope chartstroke-dashoffset

Maps to: “Migrating a retail BI estate to Snowflake”

Audit, then migrate

The BI estate audited by real usage: dead datasets shed in place, the rest carried across the Redshift → Snowflake divide, each landing with a verified-parity tick.

grid transformsstagger

Maps to: “Catching product-feed problems before they cost a month”

Feed health, caught today

A daily scan sweeps the catalogue; a product flips to disapproved and an alert fires the same day — not at month-end after weeks of wasted spend.

status gridscan sweeptoast

Maps to: “An ad-management platform with a human in the loop”

System proposes, human disposes

A change proposal travels the pipeline and pauses at the approval gate until a human ✓ stamps it — then applies, with a loop back to auto-re-check in two weeks.

animateMotion holdkeyTimes

Maps to: “Merchant analytics for a multi-brand loyalty platform”

One definition → every brand

A version-controlled dashboard template stamps itself out into per-merchant dashboards, each in its own hue — asset bundles re-deployed by script, not rebuilt by hand.

template pop-inper-brand var

Maps to: cohort analysis (store-impact · edtech)

Retention by cohort

A retention heatmap washing in on a diagonal, each cell's colour driven by one custom property — the cohort view behind the store-impact and edtech work.

alpha heatmapdiagonal reveal

Maps to: “after the fix, revenue did this” moments

Self-drawing insight line

A revenue line that draws itself in, area fill wiping behind it, a dot riding the exact curve — the headline lands only once the line arrives.

stroke-dashoffsetclip-path wipeanimateMotion

Maps to: the UK/US Shopify prospect radar

Prospect radar

A rotating beam sweeps and prospects light up as blips as it passes, qualified ones held brighter in teal — the growth engine, made visible.

conic sweepstaggered blips

No JS1.5 KB gzip

Comparison & change — who moved, against what

Before/after, gaps, ranks and profiles. The common trick: keep every entity's own line or row, so nothing can hide inside an average.

Slopegraph, before → after

Cost per workload across the migration — every workload keeps its own line, and the one that got more expensive stays amber, on the record.

Maps to: “Migrating a retail BI estate to Snowflake”

line drawgutter labels

Maps to: geo / catchment routing work

Dumbbell gap

Straight-line vs road distance per suburb — the connector makes the gap the hero, and the mountain-detour row goes amber. Why 17% of customers flip nearest-store.

connector growdot pop

Maps to: store-impact · commission replay

Diverging bars

Per-store lift and give-back from a shared zero — gains teal, losses amber, sorted so the ranking is free. Everyone named before the decision.

scaleX from zerodual origin

Maps to: merchandising / category reviews

Bump chart

Category ranks over four quarters — the “who overtook whom” story a revenue line hides. Only the climber gets the accent.

rank lanesline draw

Maps to: exec dashboards (dashboards-as-code)

Bullet KPIs

The grown-up gauge: measure vs target vs last year in one slim row, with qualitative bands behind. Three of these replace a wall of dials.

layered trackstarget tick

Maps to: maturity assessments / retainer reviews

Radar, month 0 vs 6

Data-maturity profile at engagement start and end — same five axes, same scale, and the gap between the polygons is the engagement.

polygon drawpentagon rings

Maps to: cashflow memos / monthly close

Composition & flow — what the whole is made of

Parts of a whole, and how volume moves between states. Area and width are the honest encodings here; the waffle keeps it human.

Waterfall bridge

Opening cash to closing cash with every step named — inflows teal, outflows amber, dashed connectors carrying the running level.

scaleY staggerlevel connectors

Maps to: ecommerce conversion diagnostics

Funnel with the leak named

Session → paid, but the finding is the DROP between stages: the worst step lights amber with its diagnosis, not just its percentage.

center scaleXdrop callouts

Maps to: the dp_attr first-party attribution cookie

Attribution flow

Channels → outcomes with ribbon width as lead volume — and the insight riding the thinnest ribbon: referrals are few but convert at 70%.

bézier ribbonsdash reveal

Maps to: plain-language exec reporting

Per 100 customers

Every square is a person: 76 unchanged, 15 buying more, 9 shifting channel. The two-audiences rule as a visual — the same econometrics, told for a board.

waffle gridgroup stagger

12 lines JS1.7 KB gzip

Stacked area mix

The ads layer visibly thickens while the total stays calm — the mix story a single revenue line buries. Layers share exact boundary vertices, so the stack can't gap.

Maps to: channel-mix analysis

clip-path wipeshared vertices

Maps to: catalogue / SKU concentration

Pareto 80/20

Ranked SKU bars plus a cumulative line, with the amber guide dropping onto the exact bar where 80% of revenue is reached. A budget decision, drawn.

bars + cum. lineguide drop

Maps to: merchandising / revenue composition

Treemap, area-true

Revenue by category where area is the encoding — six hand-squarified blocks, the one under review in accent. For when the parts are many and unequal.

rect pop-inhand squarified

Maps to: red-team default: small-n event studies

Uncertainty & rigour — the red-team, drawn

The standing methodology as visuals: intervals, jackknives, placebo distributions, detectability bounds. These are the charts that make a claim survivable.

Jackknife, leave-one-out

Drop each store in turn and re-estimate: every range clears zero, so no single store carries the claim. Small-n honesty, drawn instead of asserted.

whisker growpooled diamond

Maps to: red-team: “test, don't debate”

Placebo distribution

The estimator rerun on 800 windows where nothing happened — and the observed effect standing far outside what luck alone produces. A p-value as a picture.

bell path drawobserved drop-in

Maps to: red-team default: effect ≈ 0 claims

Effects with detectability

Five campaign effects with 95% intervals — the ones crossing zero go grey AND carry their power bound, so “no effect” can never quietly mean “no power”.

CI whiskersverdict colour

Maps to: cashflow scenario engine (finance)

Forecast fan

History draws, then the forecast fans out in nested 50/80% bands — width is the honesty, and the decision rides the floor: the worst decile clears zero.

nested bandsshared fork vertex

Maps to: geo / postcode uncertainty work

Histogram with a threshold

Postcode-centroid error: most codes are fine, but a real tail sits beyond ±20 km — recoloured amber and shipped as an uncertainty column, never averaged away.

scaleY barstail recolour

Maps to: decision modelling / price elasticity

Fit with residuals

Dots first (the evidence), then the elasticity line draws through (the model), then grey residual sticks admit what it doesn't explain. Order of appearance = order of trust.

dot→line→residualline draw

Maps to: store-impact per-store trends

Time, monitoring & ops — catching it the same day

Trends, seasons, breaches and progress. The recurring sell: the alert fires the day of the change, not at month-end.

Small-multiples sparklines

Eight stores on one shared scale — only the new store's catchment breaks trend, and it's the only card framed. Small multiples beat one crowded chart.

shared y-scalespark draw

Maps to: the GMC feed monitor (timeline view of 10)

Anomaly vs its band

The metric lives inside its expected band until it doesn't — one breach, one pulsing dot, one toast stamped 07:15, day of the change.

expected bandbreach pulse

Maps to: data-quality ops / exception rates

Control chart

The anti-panic tool: points inside the 3σ limits are weather, the one breach is climate. Stops teams chasing noise; makes the real signal unmissable.

σ limitsbreach blink

Maps to: retail order patterns / ops planning

Calendar heatmap

Weekday rows × week columns keep time human: weekends run hot, and Black Friday week is visibly a different animal. A staffing decision at a glance.

alpha cellsdiagonal wash

18 lines JS1.9 KB gzip

Cumulative steps

Money recovered pass by pass — recovery arrives in lumps and the step chart refuses to smooth that lie. Each riser is a named pass; the flag is R2.35m.

Maps to: the reconciliation engagement

step path drawriser dots

Maps to: the Snowflake migration plan

Delivery timeline

Twelve weeks, four overlapping phases, one diamond on the cutover — the overlaps are the design: verification shadows migration, it doesn't follow it.

bar sequencemilestone pop

Maps to: audit → roadmap engagements

Decisions & geography — the call, drawn

Where the analysis meets the decision: prioritise, choose, break even, and put customers on a map that routes like a car, not a ruler.

Effort × impact quadrants

Twelve audit findings placed by measured impact and estimated effort — the chart IS the prioritisation meeting, with the two quick wins ringed.

quadrant guidesring pulse

Maps to: decision modelling engagements

Decision tree with EV

A pricing call made in rands: branches carry probabilities, leaves carry outcomes, the chosen path draws in accent — and the road not taken stays visible, dim.

branch draw orderEV badges

Maps to: automation ROI (recon engine, monitors)

Break-even crossing

Manual cost keeps climbing; the build starts higher and flattens. Where they cross is the only number the buyer needs — and the setup premium is drawn, not footnoted.

two-line drawshared-vertex wedge

Maps to: geo / store-catchment routing

Catchment dot map

Two stores, customers coloured by nearest store by road — the amber-ringed cluster is closer to A on a map but drives to B around the bay. No geodata, just geometry.

abstract dot mapbarrier arc

Maps to: edtech survey work (parents / coaches)

Likert, diverging

Survey rows anchored on the same agree/disagree boundary so they compare instantly — and the one weak statement can't hide inside a stacked 100% bar.

outward growstraddled neutral

Maps to: fulfilment / delivery SLAs

Distributions & shape — what averages bury

Box, violin, ridgeline, beeswarm, butterfly, cycle: six ways to show the spread and the tail instead of the mean. The recurring sell: the SLA, the fairness question and the trend all live in the shape.

Box & whisker

Four carriers, near-identical averages, very different tails — and the SLA is set by the tail. The one outlier is a named incident, not noise.

box growwhisker draw

No JS2.2 KB gzip

Violin distributions

The store channel is bimodal — gifting spike, big-ticket spike — so its “average basket” is a value almost nobody spends. Width IS density.

Maps to: order-value analysis by channel

mirrored pathclip wipe

Maps to: staffing / ops planning

Ridgeline, hour × day

Seven overlapping ridges on one scale: weekdays twin-peak at lunch and evening, the weekend is one broad midday wave — a different roster, visibly.

area ridgesstagger rise

Maps to: “Redesigning a sales commission scheme”

Beeswarm

Every rep's pay change under the candidate scheme is one dot — the six give-backs stay visible people, not a footnote inside a mean.

dot swarmpop stagger

Maps to: omni-channel strategy (store-impact)

Butterfly pyramid

In-store grows left, online grows right, age bands down the spine — the channel crossover at 45 becomes a place on the chart, not a sentence.

mirrored scaleXdual origin

Maps to: retail seasonality / staffing

Cycle plot

One panel per weekday, four weeks deep inside each: Friday and Saturday share a mean, but Friday is flat and Saturday is compounding +8% a week.

panel seriespanel means

Maps to: regional / channel strategy

Composition, hierarchy & maps — parts of the whole, placed

Marimekko, sunburst, packed bubbles, tile cartogram, rose: area does the arguing, and each one carries its honesty note (√ radii, equal tiles) in the file.

Marimekko

Column width = province share, split = channel mix: “W Cape is majority-online AND a quarter of the business” lands in one shape.

width-true columnsstacked pop

Maps to: revenue hierarchy (merchandising)

Sunburst

Category ring inside, subcategories outside, every arc a single dasharray circle — the whole hierarchy is dash arithmetic, no path math.

dasharray arcspathLength=100

Maps to: subcategory revenue at a glance

Packed bubbles

Tableau's signature view, hand-packed: radius = √share so area stays honest, and the long tail gets one collective label instead of eight unreadable ones.

hand packingpop stagger

Maps to: provincial reviews

Tile cartogram

Nine equal tiles in rough SA geography — a real map would drown Gauteng's 34% in the Karoo's empty space. Equal tiles let colour carry the number.

grid mapalpha encoding

Maps to: seasonality, stock & staffing

Rose (coxcomb)

Twelve wedges around the year clock, radius = √orders so area stays honest — used because December genuinely sits next to January's trough.

sector paths√ radii

Maps to: basket / cross-shopping analysis

Relationships & structure — what moves together

Chord, hexbin, parallel coordinates, network, dendrogram, SPLOM, correlation, connected scatter. The correlation family nominates hypotheses; the causal charts (06, 73) get to promote them.

Chord diagram

Ribbon width = customers shopping both categories; the bundling decision rides the fattest ribbon. Kitchen ⇄ Home carries 31% repeat traffic.

bézier ribbonsdasharray arcs

Maps to: geo / catchment density

Hexbin density

599K customers → 1,798 postcode points → 37 honest bins, one polygon stamped with <use>. Density without the overplotted dot soup.

<use> gridalpha bins

Maps to: store portfolio reviews

Parallel coordinates

Six stores, five shared axes, one shape each — and the airport store inverts on every axis. A per-metric averages table calls this portfolio “fine”.

polyline profilesoutlier accent

Maps to: systems integration (recon, warehouse)

Integration network

The retired point-to-point mesh stays faintly visible under the six governed spokes — the argument is the contrast, not just the tidy end-state.

marching dashesnode pop

Maps to: store clustering / segmentation

Dendrogram

Stores merge right-to-left in similarity order; the dashed cut turns 8 stores into 3 archetypes — and the merge heights show how different they really are.

elbow pathsmerge-order draw

Maps to: CRM / loyalty metric trade-offs

Scatterplot matrix

Every pairwise relationship at once — six panels from three dot clouds, because each mirror cell is the same <g> through a transpose matrix.

transpose reusefinding frame

Maps to: channel & pricing diagnostics

Correlation matrix

Teal positive, amber negative, opacity = |r| — with the values printed, because a heatmap without numbers is a mood board. The finding: AOV ↔ Returns.

diverging alphadiagonal fade

Maps to: pricing / elasticity work

Connected scatter

Price vs volume with time as the thread: three rises cost nothing, the fourth broke volume — the elasticity cliff has an address on the path.

dated path drawamber break

Maps to: migration / engagement health

Time, finance & motion — density and drama, earned

Streamgraph, horizon, candlestick, bar race: the dense forms for walls of context, and the one animated format that earns its drama because the overtake is the finding.

Streamgraph

Ticket mix flows around a centre axis: feed & data queries swell after go-live while total thickness barely moves — a training gap, not an incident.

shared verticesclip wipe

Maps to: ops monitoring walls

Horizon strips

Six metrics × 36 weeks in 240 vertical px: values past the cap re-draw darker on top, negatives mirror up in amber. The two incidents are unmissable.

fold overlaymirrored negatives

Maps to: cashflow memos / treasury

Candlestick cash

Weekly OHLC candles for the cash balance: every Friday close said “fine”, the wicks show two weeks kissed the overdraft mid-week. Range is the risk.

wick + bodyfacility line

Maps to: category reviews (static twin: 21)

Bar race, no JS

Smart home climbs rank 5 → 1 across four quarters — lane hops and bar lengths on synced CSS keyframes with plateau holds, playing forward and back forever.

keyframe holdslane hops

Maps to: trading / promo reviews

Boardroom & evaluation — decisions and models, audited

Combo, table lens, KPI cards, lift, survival, event study, tornado: what execs actually read, plus the charts that keep models and causal claims honest.

Dual-axis combo

Revenue bars, margin line: November's record bar sits on the line's floor — the promo bought revenue at 14% margin. Both axes zero-anchored and labelled.

bars then linehonest axes

Maps to: exec reporting (dashboards-as-code)

Table lens

Execs trust tables and skim charts — in-cell bars and delta chips give them both: exact rands stay auditable, magnitudes become impossible to misread.

CSS grid tabledelta chips

Maps to: dashboard headers

KPI cards

The BAN done properly: count-up numbers, sparkline context, and chips coloured by good/bad rather than up/down — falling returns stay teal.

count-upsemantic chips

14 lines JS2.2 KB gzip

Lift curve

The random-baseline diagonal draws FIRST — house rule: no score ships without beating the trivial baseline. Top 20% of the list captures 58% of conversions.

Maps to: red-team default: trivial baselines

baseline firstgain wedge

Maps to: edtech / subscription retention

Survival curves

Kaplan–Meier retention with censor ticks: recent joiners are counted for the months we saw them, never guessed beyond. Onboarding is +27pt at month 24.

step pathscensor ticks

Maps to: store-impact DiD (red-team)

Event study

Six pre-months of coefficients hugging zero, then the lift arrives and persists — parallel trends shown, not assumed. The chart the diff-in-diff (06) stands on.

CI whiskerspre/post split

Maps to: decision modelling (feeds 39)

Tornado sensitivity

Each parameter swings the ROI while the rest hold at base, sorted by swing: adoption moves the case 3× more than build cost — de-risk that first.

centre-anchored barssorted swing

Best for: exec narratives where every inflection has a reason to act on

Comparison & change, encored — same order, louder point

Charts you already know, re-cut so one thing lands harder — the turning point, the endpoint, the spread, the shift, the gap. Hover a card for ✎ Edit.

Annotated line

A trend line where the labelled turning points carry the story — each bend tagged with its cause. The line is context; the callouts are the argument.

line + areacausal callouts

Best for: long ranked lists where full bars feel heavy or moiré

Lollipop ranking

A bar with the ink stripped out — thin stem, weighted dot — so the eye lands on the value, not the mass. Same order, far less clutter.

rankedstem + dot

Best for: two states across many categories where the change is the message

Cleveland dot plot

Two dots joined by a stick; the gap between them is the improvement, read without subtraction. Cleaner than paired bars, which double the ink.

before / afterconnector = delta

Best for: naturally cyclical categories (hours, months, compass) or a punchy hero

Radial bars

24 hours bent into a clock, so cyclical time becomes position — midnight meets midnight and the two daily peaks sit where intuition expects them.

polartiny JS gen

SVG · tiny JS1.8 KB gzip

Range bars

A floating bar from fastest to slowest with the median marked — the spread an average would hide, and the tail that makes a customer angry.

Best for: categories where spread matters as much as the centre — SLAs, price ranges

min–median–maxrisk flag

SVG · no JS2.1 KB gzip

Mix-shift ribbons

Two 100% stacked bars linked by ribbons that show the flow — how each channel's share redistributed, not merely that it changed size.

Best for: how a composition moved between two moments — channel, budget, portfolio

100% stackedflow ribbons

Best for: explaining a gap between two totals — budget vs actual, plan vs forecast

Variance bridge

A waterfall that walks from budget to actual one driver at a time, decomposing a scary total into causes you can act on — and proving it lands on the endpoint.

waterfallrunning total

Best for: one variable's shape, compactly, where a full histogram is overkill

Distribution & rigour, extended — the shape and the doubt

The red-team habit, drawn: show the whole distribution and the uncertainty, not a single point that pretends to certainty.

Density strip

A density curve with the raw rug beneath and both median and mean marked — the gap between them is exactly the skew an "average" buries.

density + rugmedian vs mean

Best for: checking a distributional assumption before you trust a model or t-test

Q–Q plot

Sample quantiles against a normal line; here the points lift off in the upper tail — a fat right tail that would make naive confidence intervals too narrow.

quantile–quantilenormality

Best for: SLA and percentile questions — P50 / P90, "what share is below X"

ECDF with percentiles

An empirical CDF — every step a real order, so any percentile reads straight off the axis. No binning choices to argue about, unlike a histogram.

step curveP50 / P90 guides

Best for: agreement between two systems measuring the same thing — SAP vs bank

Bland–Altman agreement

Difference vs mean with limits of agreement — two points breach them, both mis-posted to VAT clearing. The check that catches a right amount in the wrong account.

difference vs meanlimits of agreement

SVG · no JS2.1 KB gzip

Marginal scatter

A scatter with a histogram on each margin — the relationship and both distributions in one frame, so a correlation driven by a clump can't hide.

Best for: when the relationship and each variable's own shape both matter

scatter + marginalstiny JS gen

SVG · tiny JS2.2 KB gzip

Forest plot

Each segment's estimate with its CI against a no-effect line; whiskers that cross zero are greyed, and a pooled diamond gives the honest bottom line.

Best for: comparing an effect across subgroups without cherry-picking

effect + CIpooled diamond

SVG · no JS2.2 KB gzip

100

Uplift with CI

Test-vs-control lift bars topped with confidence intervals; the one whose interval straddles zero is noise you shouldn't bank.

Best for: experiment / incrementality results where "beyond chance?" is the question

A/B lifterror bars

Best for: tracking delivery against a plan — sprints, migrations, close checklists

Time, flow & monitoring — pace, pipes and staleness

Operational visuals for catching it the same day: are we on pace, where's the jam, and which feed just went stale.

Burndown

Remaining work against the straight ideal line; ride above it and you're behind, and the shaded gap is exactly how far.

ideal vs actualbehind-gap

Best for: flow systems where work moves through stages — pipelines, queues, WIP

Cumulative flow

Stacked stage bands over time; a widening band is a bottleneck you can see before any table names it — here, the qualify stage.

stacked bandstiny JS gen

SVG · tiny JS2.1 KB gzip

Punchcard

Weekday × hour, each cell a dot sized by volume, so empty hours read as genuinely empty. The staffing question answered at a glance.

Best for: patterns across two cyclical keys where magnitude varies a lot

day × hoursize = volume

SVG · tiny JS2.0 KB gzip

Uptime ribbon

One bar per day, status-page style — 98% green with a lone red spike tells the reliability story faster than any number.

Best for: a compact history of a mostly-binary daily state — uptime, SLA, backups

status striptiny JS gen

SVG · tiny JS1.8 KB gzip

Deviation area

The space between actual and expected, filled and coloured by sign, so a soft dip reads as a shape you feel — not a wobble you squint at.

Best for: emphasising departures from a reference — vs budget, vs seasonal norm

signed fillline + area

SVG · no JS1.7 KB gzip

105

Freshness dots

Each source a dot on an "hours since update" track over green / amber / red zones — the bank feed at 26h and SARS at 70h are past their SLA.

Best for: monitoring recency across many feeds at once — pipelines, dashboards, backups

stalenessSLA zones

Best for: a strict part-of-whole hierarchy read as levels, not rings

Structure, decisions & finance — the call, the money, the data

Hierarchies and relationships placed to read; the trade-off and the money made explicit; the source profiled before you trust it.

Icicle hierarchy

A P&L read top-down as nested rectangular levels — fairer for small slices than a sunburst's wedges, and every level stays a scannable row.

partitionlevels

Best for: pairwise relationships over an ordered or small set of nodes

Arc diagram

Categories on one line, an arc per co-purchase; a stable, readable layout where a force-directed graph would wander into a hairball.

one-axis networkarc = strength

Best for: dense pairwise data where a node-link view overlaps itself

Adjacency matrix

Co-purchase as a grid of sized dots; order the axes and the communities appear as bright blocks a network diagram would tangle into a hairball.

matrixreorder → blocks

SVG · tiny JS1.9 KB gzip

Proportional symbols

Bubbles sized by count on a schematic map, so a huge empty province can't shout as loud as a small dense one the way a choropleth would.

Best for: mapping a count across regions where area would mislead

bubble maparea = value

SVG · no JS1.7 KB gzip

Decision matrix

Options × weighted criteria as a red-amber-green table with a computed score, so "why not the cheap one?" has an auditable answer.

Best for: a defensible multi-criteria choice — vendor, tool, prioritisation

RAG tableweighted score

SVG · tiny JS2.2 KB gzip

Cost curve

Savings ranked cheapest-first, width = size and height = cost; everything left of the pay-back line funds itself before the expensive rebuild is worth a meeting.

Best for: prioritising a portfolio of actions by cost and size at once

width × heightpay-back line

Best for: any "how long have we got?" — cash runway, inventory cover, headroom

101

Cash runway

Balance projected to its zero-crossing at the current burn, with an act-by marker three months earlier — the real deadline, because decisions take time.

line + areaact-by marker

Best for: a split-by-attribute that carries a risk reading — currency, fixed vs variable

102

FX exposure

Each outflow stream split into its USD/GBP and ZAR parts, with a weighted headline — the foreign exposure to hedge.

split barsexposure %